Describing AI

Companies are investing millions of pounds to develop Artificial Intelligence (AI) technology. Many people use that AI technology daily to make their lives easier. But search Google for images of “Artificial Intelligence”, and you’ll be faced with a sea of glowing, bright blue, connected brains. The imagery used to illustrate AI is a far cry from the more mundane reality of how AI looks to its users, where it powers services like navigation, voice assistants and face recognition. The disconnect makes it hard to grasp the reality of AI.

AI looks more like this, to its users (photo credits: Pixabay and Wikipedia)

The language used to describe AI can be as problematic as the images used. With AI increasingly impacting so many people’s daily lives, it’s important to represent it accurately and not to obscure the responsibility of those designing the system to get to right.

‘The AI taught itself”

Most articles about AI are talking about machine learning (ML). That’s a set of algorithms that train models from data. After training, the models encode patterns that exist in the training data. Machine learning is the best way we know for computers to carry out complex tasks, like understanding speech, which we can’t explicitly program them to do. 

We talk about ML models learning, so it’s tempting to write that “the AI taught itself to…” do something. For example, AI has taught itself to prefer male candidates, to play games, to solve Rubik’s cube, and much more. The phrasing “taught itself” is problematic as it suggests the computer and the machine learning model has agency. It hides the reality of machine learning, where a scientist or engineer carefully constructs both the dataset and model, then spends significant time building and tuning models until they perform well. The model only learns from the training data, which is usually constrained in some way. Perhaps there’s only a limited amount of data available, or limited time and computing power in which to run the training. Whatever the constraints, what the model ultimately learns is strongly influenced by the design choices of its creators.

“The AI cheated”

Bots also teach themselves to cheat, perhaps by hiding data or exploiting bugs in computer games. Yet, ‘cheat‘ is an emotive word which also suggests there’s intent behind the cheating.

Cheat (verb): “act dishonestly or unfairly in order to gain an advantage”

The task that an ML model is trained to do is well defined by its creator. Often, imperfect simulations or video games are used to mimic the real-world, because it’s too expensive or dangerous to be outside. In these simulations, computers can repeat the same action many more times than people, and act a lot faster, so they can discover and exploit bugs that people can’t. They’re not slowed down by the physical act of clicking the mouse or pressing a button. Uncovering and exploiting bugs in the task is a perfectly reasonably outcome to expect of a machine learning model. We have a tacit understanding that exploiting bugs in a simulation or computer game is cheating. Computers, however, don’t know that this isn’t within the spirit of the game. They are blind to the difference between a rule of the game and a bug in its implementation.

“The bot defied the laws of Physics!”

In the real world, nothing can defy the laws of Physics. When a simulation defies the laws of Physics, something is wrong. Machine learning models can exploit bugs in the simulation to invent behaviour which we intuitively know is wrong but which satisfies the constraints of the task. Still, the results can be entertaining and tell us something about how machine learning works:

 “When they gave the algorithm no constraints and asked it to cross an obstacle-laden course, the AI built an extremely tall bipedal creature that simply fell down to reach the exit.”

New Scientist

“AI can tell if you’re a criminal”

Published research has suggested that AI can tell if you’re a criminal or your sexual orientation from a picture of your face, tell if you’re lying from your voice, or identify your emotions from your gait. But, even the best performing AI cannot do the impossible.

The datasets used for training machine learning models have to be accurately labelled. If people cannot accurately label a dataset, then it’s very unlikely a machine can learn the task. Sometimes it takes an expert to label a dataset – almost anyone can transcribe audio but very few have the expertise to label MRI scans – yet accurate data labels are essential to a machine learning task. Correctly labelling sexual orientation, whether you’re lying, whether you’re a criminal, or what emotional state you’re in is an impossible job for anyone.

There are some scenarios where machine learning can uncover patterns which people cannot see. For example, MRI scans can be labelled using additional information from biopsies. Perhaps the doctor doing the labelling isn’t sure from the image whether a mass is cancerous, but additional information from a biopsy can help with labelling the image. In this scenario however, you have to take extra care with the setup and evaluation to make sure the model is indeed learning something new and novel.

Another point to note about the datasets used for training ML models is that they are constructed by people working within constraints. Those constraints can be related to time, money, labelling effort or something else entirely. For example, in the study of sexual orientation, the researchers scraped profile photos from dating websites. Dating profile pictures are chosen by the users from amongst many pictures of themselves to convey a particular image to the world. Predicting sexual orientation from these photos tells us more about cultural norms of gender presentation than about whether you can predict someone’s sexual orientation from their face*. Sometimes, an impressive result on a task in a lab environment does not transfer to the real world.

Machine learning and AI have shown some impressive results recently – including translating between languages, filtering your spam email away, answering questions, detecting fraud, driving cars and more. Taking away the hyperbole doesn’t detract from its successes, and ultimately will make it easier for everyone to understand.

---

* NB: you can’t, and shouldn’t. For a deeper dive on the sexual orientation study, see this article

What is Artificial Intelligence?

Artificial Intelligence (AI) is a fast growing field. The 2018 AI Index report illustrates just how fast it is growing. It reports that published research papers in AI have increased 7x since 1996, university enrolment on AI courses has increased 5x since 2012, investment in AI startups in the US has increased 113% since 2012 and mentions of AI and machine learning (ML) in the earnings calls of tech companies have increased more than 100x since 2012. These statistics show how AI is growing not just in academia, but the technology is rapidly being adopted by businesses and becoming commercialised.

Yet, AI is an ambiguous term which has no well defined and agreed upon definition. It’s typically used as an umbrella term covering a variety of techniques that make computers appear to have human-like intelligence. Recent advances in AI have largely been driven by machine learning – a set of algorithms which learn their behaviour from data.

The Beginning of AI

The term ‘artificial intelligence’ was chosen by John McCarthy for a workshop to be held in the summer of 1955 at Dartmouth College. The workshop brought together several leaders in computing and had ambitious goals:

“An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves. We think that a significant advance can be made in one or more of these problems if a carefully selected group of scientists work on it together for a summer” – Dartmouth workshop proposal

Since this 1955 workshop, there have been several periods of excitement around AI. As a result, when the technology didn’t live up to the hype, these were followed by a drop in interest. There were two notable periods of lower investment in the technology (1974–1980 and 1987–1993) nicknamed ‘AI winters‘. While the Dartmouth College workshop didn’t solve the problems it set out to, it set the direction for AI research in the next decades.

Expert Systems

By the 1980s, expert systems were the most popular approach to AI. Expert systems encode human knowledge about a domain as rules, and allow inference over that knowledge. For example, suppose we have rules stating that the Shanghai tower is taller than the Empire State building and that the Empire State building is taller than the Eiffel Tower. From these, the expert system can infer that the Shanghai tower is taller than the Eiffel Tower using logical rules about height.

Expert systems ultimately became hard to maintain, brittle and difficult to work with. However, knowledge bases, like Google’s Knowledge Graph and Amazon’s Alexa knowledge base, still form part of modern AI systems. They encode knowledge about the world that these virtual assistants rely on to answer questions.

Machine Learning

Machine learning (ML) refers to a set of algorithms which learn their behaviour from data. It’s been around for decades, but became more popular from the 1990s onwards. A lot of the current discussion about AI concerns technology which is based on supervised machine learning methods, in particular deep learning and neural networks. These have been very successful in the past 5-10 years.

Consider automatic speech recognition (ASR) – the task of automatically transcribing human speech. It’s simply not possible to write down the rules, or knowledge, to explicitly program a computer to do this task. Human speech is highly variable, and even the same person cannot repeatedly speak exactly the same utterance in exactly the same way.

For these reasons, machine learning is used as the basis of ASR systems which learn how to transcribe speech from examples of manually transcribed audio. Modern systems might use thousands of hours of audio and tens of millions of words of text to learn from.

There are three broad sub-categories of machine learning.

Supervised learning is where a computer learns how to do a task from labelled examples:

• Speech recognition systems are trained from labelled audio data
• Object recognition systems are trained from labelled photos – ImageNet is a dataset of labelled images containing more than 14 million examples across more than 20,000 categories.
• Spam email classification uses examples of spam email to learn from

Unsupervised learning learns patterns, groups and categories in unlabelled data:

• Clustering a set of scientific papers to uncover the topics they are written about (topic modelling with LDA)
• Clustering segments of audio to discover how many speakers are represented in the set

Reinforcement learning is used in scenarios where the machine has to take a number of steps in an uncertain environment, towards a goal, before it can know whether those actions were good:

• Having a multi-turn dialogue with a person to complete a task like booking a restaurant table
• A robot traveling to a particular location when it can’t sense its environment perfectly

Machine learning has been hugely successful in recent years due to a) increased data, b) increased computation, and c) improved algorithms. With cheaper storage and connectivity, data can be more easily collected and shared. Increases in computing power and the introduction of cloud computing mean we can quickly train on larger and larger amounts of data. Cloud computing and smartphones mean we can offload intensive computation to the cloud while people access AI through their phones. Neural networks/deep learning algorithms have their roots as far back as the 1940s, but with increased data and computational power have become very powerful machine learning techniques.

Popularity of ‘deep learning’ – Google trends

Though recent performance improvements have been impressive, no machine learning algorithm is perfect. We always expect some error in the output. A key measure of system performance is its error rate.

A word of caution

Despite recent advances, AI and ML technology still need care and awareness so they are built in ways which do not cause harm and discriminate, particularly when being deployed at scale.

Not all tasks are able to be learnt by machine learning. Some tasks are impossible, like identifying criminality or sexuality from your face. A good rule of thumb is that a person has to be able to do a task in a couple of seconds with ease, otherwise you cannot train a machine to do that task.

When it is possible to learn from data, machine learning uses a finite dataset. If a system is built on a training dataset and then deployed in a scenario that’s different from the training data, it doesn’t perform as well. For example, an object recognition system will not be able to recognise images of fish if it’s only ever been trained on images of plants. This becomes much more complex if, say, your object recognition system is trained on images from the US and used in the real world on images from Asia. Now the technology which works well to identify objects in the West doesn’t work well in other regions of the world. Along similar lines, facial recognition can perform poorly for certain segments of the population if it’s not trained on representative data.

Another issue with learning from data arises when ML is seeking to automate human decisions which are already biased. In this situation, human bias is embedded in the training data set. Identifying potential reoffenders using data in an unbiased way is tough because any dataset collected to train the system already contains real-world bias. Any AI system would replicate that bias.

There are many more concerns with adoption of AI technology. Fairness, bias and transparency are an active area of research e.g. see the recent NeurIPS workshop and FAT/ML conference.

Conclusions

Artificial Intelligence is behind many of the products we use today – virtual assistants, spam email filtering, fraud detection, online ad placement and much more. It is being used daily by large numbers of people around the world.

The success of AI has been driven by other technological improvements such as data, connectivity, storage, cloud computing and smartphones, as well as improvements in the core technology. On many tasks, deep learning with neural networks has drastically improved performance in recent years.

Yet, AI is not a silver bullet. It has behaviour and limitations which must be understood by both those building and those using the technology.

London Tech Week 2019

London Tech Week was held 10-14th June this year, and hosted many events including two I spoke at: CogX and AI Summit.

Despite the heavy rain, CogX seemed busier than last year. I was impressed at the diversity of both speakers and the range of topics they talked about. Sadiq Khan opened the event by talking about how London has one of the one of the world’s leading tech sectors, including over 750 AI companies, and is at the forefront of this industry.

On Tuesday morning I spoke on the Lab to Live stage about building custom voice technology – the talk is HERE (from 43:25). In the same session, I loved hearing about how Tom Hewitson and Jess Williams built their businesses on top of Alexa Skills. Kudos to Julian Harris for putting together such a fantastic range of speakers on conversational AI.

Later the same day I, with Libby Kinsey, curated a session ‘Beyond Voice Assistants’ on the Cutting Edge stage. You can find the recording HERE. We brought together a panel of experts with deep knowledge of building and researching voice technology and discussed some of the directions that might emerge in the next few years:

• Professor Verena Rieser at Heriott-Watt University
• Dr Kate Knill from Cambridge University
• Grace Boswood COO of the BBC’s design and engineering department

 

On Wednesday and Thursday I headed to the ExCeL centre for AI Summit. Here I spoke on stage again about building custom voice technology, and also took part in a panel about the pros and cons of outsourcing AI. The panel had different opinions on this question and there’s no right answer, as each business has different motivations and constraints they have to work within. Outsourcing and partnering with experts can be a great way to start building AI, but businesses have to be careful in the process not to outsource core business functions.

In both conferences, it is encouraging to see that ethics of technology and its impact on our society was higher on the agenda than in previous years. This reflects the changing media narrative around technology. Sadiq Khan succinctly summed it up at the end of his address:

“It must ultimately fall to government working with tech businesses and leaders to ensure that AI adoption is always steered towards augmenting, not replacing, human thought and endeavour. That citizens will always be at the heart of AI design, and that the public can be reassured that AI will not lead us to some dystopian future but to a better one for us all.” – Sadiq Khan

 

The AI Gender Gap

In the past few years, machine learning (ML) has become commercially successful and AI firmly established as a field. With its success, more attention is being paid specifically to the gender gap in AI. Compared to the general population, men are overrepresented in technology. While this has been the case for several decades, the opposite was true in the early days of computing when programming was considered a woman’s job.

Diversity has been shown to lead to good business outcomes like improved revenue. It’s also important that diverse voices are represented in the design of AI products which are used by large numbers of people.

Women in AI

Understanding the situation in the field of AI requires data. However it is hard to measure precisely the gender skew in AI as the workforce is global, fast growing, and highly mobile.

There are two recent studies estimating the gender gap in AI. Element AI’s Global talent report 2019 looks at who publishes in major AI research conferences and finds that 18% are female. Their 2018 report found 12% female participation. The second report – the World Economic Forum’s Gender Gap Report – found that 22% of the AI workforce is female.

Few companies publicise the proportion of female research staff they have working specifically on AI. A WIRED article found from their public research profiles that 10% of Google’s research workforce is female, and 15% of Facebook’s. Research Scientist is an important job category in industry as these roles are often the most coveted and highly paid.

These numbers are in line with the overall proportion of women working in technical roles in the UK. The IET currently estimate that 11% of the UK Engineering & technical workforce are female and the BCS estimate 17% of the UK IT workforce are women.

Career Progression

The overall numbers are not broken down by seniority. Research from the Anita Borg institute showed a 50% drop in women’s participation between entry and executive levels across the US technology industry. Similarly, the WEF report states:

“Male AI professionals are better represented in roles such as software engineer, head of engineering, head of IT as well as business owner and chief executive officer—positions that are generally more lucrative and of a more senior level.”

– WEF Gender Gap Report

A separate report from Inclusive Boards found that in the UK:

“Almost two-thirds (65%) of boards in the top tech firms had no female directors. Over two-fifths of executive teams in the top tech firms had no female representation” – Inclusive Boards Tech Report 2018

The same report highlights the lack of overall diversity, with Black, Asian and Minority Ethnic (BAME) people making up just 8.5% of senior leaders in the UK tech industry.

These figures demonstrate how the proportion of women and minorities drops further at senior levels. This is important as senior employees typically have greater influence over product decisions and future directions.

Founding startups is another way that AI technology is being built. According to Tech Nation, investment in UK AI companies reached $1.3bn in 2018. Yet, research from the British Business bank showed that for every £1 of VC investment in the UK, 89p goes to all-male founding teams.

The Pipeline into AI

One reason for the lack of women in technical roles is the lack of girls choosing to pursue these careers. There are many paths into AI and a variety of jobs within the field. However, the highest paying AI jobs require a technical education, including some combination of maths, physics, engineering and computer science.

In the UK, A-levels are chosen at age 16, and are the first time that students can significantly narrow the choice of subjects they study. According to the Institute of Physics, around 20% of A-level Physics students are female and the Royal Society report that A-level Computing has a 9% take-up from girls. Girls accounted for 39% of this year’s maths A levels, and 28% of further maths A levels.

The figures continue through into undergraduate study at university. 17.6% of computer science university students in 2017/18 were female and 18.2% of Engineering students.

Conclusion

Together, these statistics paint a picture where women make up less than 20% of the AI field, dropping further at senior levels. In the UK, girls choose to study science and technology subjects at a lower rate than their male peers, both in secondary school and at undergraduate level. Leaving university, women choose not to pursue technology careers at the same rate as men. Women subsequently leave the technology workforce at a faster rate than men and face stronger headwinds when progressing their careers.

There are two important threads to addressing the gender gap in AI which must be tackled in parallel. The first is increasing the numbers of girls choosing to study STEM (Science, Technology, Engineering & Maths) subjects and follow these careers. The second is to better support women already in these careers so they can progress effectively. It is easier to build up initiatives to address the first of these, but progress will remain slow unless the second is addressed.

The statistics above have illuminated the gender gap in technology and AI. However, diversity means more than simply including women. More research into other minority groups is needed to see a fuller picture of who is working on AI, and who isn’t.

 

Update (20/07/2019): A couple of days after I wrote this post, Nesta published comprehensive research looking into the percentage of women authors of arXiv papers on AI topics. They found similar results to the two existing studies; 13.83 per cent of the authors are women. They also looked deeper at large tech companies publishing on arXiv where the ratio is similar: Google (11.3%), Microsoft (11.95%) and  IBM (15.66%). These figures show that the percentage of women working in the core research roles is lower than the percentage in all of AI-related roles. 

Resisting the 80 hour week

The “long hours” culture is glorified in many industries, including the tech industry where I work. Rachel Thomas writes clearly about how this is discriminatory and counter-productive, and tweeted:

For anyone in a job with a healthy work culture, consider blogging about it:
– offer lessons learned & advice for other companies
– with many high profile examples of toxic overwork, it can feel like "everyone" is doing it. Helpful to see counterexamples
– good for recruiting https://t.co/MaUnA4cjK7

— Rachel Thomas (@math_rachel) January 25, 2019

As a PhD student and in my early career, I often worked long hours. At other points in my career I’ve worked fewer hours, and sometimes not at all for several months. I have never put in anything approaching 80 hours a week for a sustained period of time, and yet I still consider myself to have had a successful career in technology.

I have two young children. Returning from maternity leave the first time round, I knew that I didn’t want to work full-time and so moved into a role which I could do 3 days a week. The role had been advertised as full-time, but I knew the team and they agreed to hire me on a part-time basis. This worked out well as I soon became pregnant for the second time. Combining pregnancy and work with a young toddler at home, I was physically very tired and happy not to be working full-time.

Heading back to work after my second maternity leave was a different story. I chose to go back into full-time work when my second child was a year old, and stayed full-time for 4 years. By the end of the 4 years both my children were in school and I started to experience the difficulties that this brings. Christine Armstrong has written extensively about the particular difficulties of combining parenthood with full-time work. School brought more additional admin and complex logistics than I anticipated – getting the right kit and uniforms, organising after-school clubs and activities, parents evenings, school trips – and in the UK school hours don’t align easily with a working day. It is hard to find childcare past 5:30pm, which is exactly the time that those on the US West Coast begin their day.

I have also taken three significant periods of time out of paid work over the last years. Maternity leave in the UK is generous, you are entitled to take up to 12 months out and return to a similar role (though sadly pregnancy discrimination is rife). I took 11 months off from paid work with my first child in 2011 and 8 months with my second in 2013. More recently, after 4 years working full-time with 2 young kids, I simply needed a break! I took almost 5 months out of work, most of the second half of 2018, to recharge and spend time with my family.

At the beginning of 2019 I began working at Cobalt Speech, again part-time, 4 days a week. As with my previous part-time position, this role came through my network and so I felt able to ask for reduced hours upfront. Working 4 days gives me time to pick up my kids from school at least one day a week, and have extra time in the week to do a bunch of things I don’t have time for when working full-time.

Not everyone is able to work fewer hours or take periods of time out. I am lucky to have experience working in AI/ML where there’s a currently shortage of talent. I was also fortunate to have good health in my early career years so I could gain experience, and to have a supportive husband who has also worked flexibly at different times. Still, I can’t say I always get the balance right!

Being able to work flexibly in different ways at different times has helped me keep my career going. I hope that more employers in the tech industry continue to make flexible and part-time working a reality for those who need it.

Light-up snowman decoration

I’ve been looking for craft projects to do with kids, and just in time for Christmas made this light-up snowman decoration. I used:

• A snowman shape with holes cut out for buttons
• 3x 3mm LEDs (mine are from here)
• 3x 100 ohm resistors (from here)
• 3V coin battery
• Tin foil
• Sellotape
• Paint
• Ribbon

I started with a snowman shape:

Ready-cut snowman

This one was lasercut at Makespace but you could easily cut one out of thick cardboard. The important thing is that the button holes are big enough to poke the 3mm LEDs through.

The next step was to find a circuit for 3 LEDs that could be powered by a small battery. I chose a 3V coin battery because I could easily stick it to the snowman and avoid wires hanging off. I also wanted a circuit I could put together easily without too much specialist equipment.

3 parallel LEDs

On the back of the snowman I arranged the circuit with the LEDs poking through the buttonholes. There wasn’t much room and getting the circuit in place was fiddly. Tin foil made tracks down either side of the snowman’s body and the battery was stuck at the base of the snowman. Solder and wire would be a better choice than tin foil, but would also make it less child-friendly.

The circuit on the snowman

I stuck the whole circuit down with sellotape, painted the front, strung some ribbon through the hat and hung the snowman on the tree:

Finished article

 

Explaining AI to primary school kids

Recently I was asked to talk about computers and artificial intelligence to year 5 and 6 primary school kids (i.e. ages 9-11). I’ve spent a lot of time explaining AI and ML to a range of different people, but never to such a young audience! I couldn’t find much inspiration online so came up with my own material. Here’s the outline of my lesson plan which had me spending 30 minutes with each class.

1. Introduction
My day job involves voice recognition, so I start by introducing myself and asking the class how many of them had ever talked to a computer or a device before to get something done – normally I get the majority of the class putting their hand up.

For some context, I include a bit of history. Primary school children have grown up around smartphones and voice computing and often don’t know how fast technology moves. At the time of writing, the Y5 class was born in 2008/2009. For comparison, the iPhone was launched in 2008 and Siri was released on it in 2011.

2. Computer Programming
The first discussion topic is how computers are programmed to do things by writing rules (or programs). It’s possible to write very complicated and seemingly intelligent computer programs this way. By Y5 and 6, the children in this school have had lessons programming using scratch. They understand how you make computer programs from instructions, and how it’s easy to get those instructions wrong so the computer doesn’t do what you meant it to.

3. Rules to identify cats
People like to share photos on the internet, and they share a lot of photos of cats! One thing that might make a computer artificially intelligent is being able to say whether a photo is of a cat or not. I brainstorm with the class for some rules to write on the board that we could use for identifying a cat in a picture. Some of the rules we’ve come up with in previous sessions are:

• Has a tail
• Is furry
• Is cute, makes people go ‘awww’
• Has whiskers
• Has eyes
• Is an animal

These aren’t rules that we can directly implement in a computer, but they get the idea across.

Next we go through a slide deck of pictures playing ‘cat or not’. My slides have a range of pictures – some are easy to identify as cats or not, some are cats which are obscured or in funny poses, and some are other animals which have some of the same features as cats. They get the children thinking about how to define the task (do big cats count as cats or not?), edge cases (what about the drawing of a cat?) and the kind of rule you really need (how exactly do you distinguish a red panda from a cat?).

4. Intelligent computers
After this, I ask if the kids thought the rules we had were good or not. Most acknowledge that our rules don’t cover all the pictures and so could have been better.

Some tasks, like seeing and hearing, are impossible to write rules for that a computer can follow. We need a different method. Instead of writing rules, we have a computer learn how to do the task from data. To do this we take a lot of data (images, audio, video etc.) and have people label it. For images the people might label each image what object is depicted. From this database of ‘labelled data’, we can train a machine to learn the patterns of what makes a picture of a cat or not. Once the computer has learnt from this training data, we can take the model that the computer builds and use it to identify cats in pictures it’s never seen before.

Identifying cats or other objects in pictures might seem frivolous, but there are lots of ways you can use this technology in the real world. One example which a lot of people are working on is to identify what is in front of you in a self-driving car. Another is helping to save endangered wildlife by identifying animals and counting them in the wild.

5. Discussion and Questions
I finish with a discussion about what kind of smart robots the class would build in the future. Where I know anything about their suggestions, I’ll talk about what’s being already built in that area.

Examples of ideas I’ve heard include:

• Computers that help blind people: examples include Microsoft’s Seeing AI or text-to-speech technology which can read text aloud
• Space travel: we have sent men to the moon and rovers to Mars already, but AI can help us explore further e.g. rover navigation or identifying objects in space,
• Looking after people in hospital: there are many ways computers can help in hospitals from surgery to diagnosis to doing admin and so freeing up the time of doctors and nurses to spend with patients.
• A robot that can cook: the sort of robotics needed to handle ingredients is tough to build, but there are examples
• A computer that does homework. I don’t normally point out that voice assistants like Alexa can already answer a lot of questions

When I ask the class teacher, they usually just want an intelligent computer to help them with marking homework!

There are a few other discussion points I add into the discussion:

• Computers aren’t always right. When computers learn from data, we always know they’ll make some mistakes.
• Some things are not possible for computers to do, e.g. tell if someone is lying. In general if it’s hard for a person to do, it’s also hard for a machine to learn.
• There are bad uses of technology as well as good ones. Sometimes people in the class come up with ideas that others find creepy, which helps illustrate the point.

Extra background
For those who are interested to know more, image recognition has a long history and had been well researched over recent years. One database of images, called ImageNet, has been the basis of much academic research on object recognition. This database has several million images in more than 20,000 categories.

In 2012, it took 16000 computers to learn how to identify a cat. In recent years, researchers have also looked at harder tasks like distinguishing between 5000 species of animal, identifying and tracking objects in video or automatically creating in-depth descriptions of what is happening in a photo.

Reading list: new to management

Over the past years I moved from an individual contributor role into a management one. I had some classroom training from my employer through the transition, but I also found a few books helpful along the way. Here are some of my favourites:

• One Minute Manager – was recommended to me by my mentor when I began to manage others, and I’ve since recommended them to other new managers too. There’s a whole series of books, and I found them a great starting point to illustrate a few basic ideas.
• Peopleware – more in-depth than the One Minute Manager, on how to manage people and projects.
• Crucial Conversations – on having difficult conversations, a part of any management role.
• The Five Dysfunctions of a team – on forming teams and understanding team dynamics.
• The Leadership Pipeline and The Manager’s Path – for putting management into context, and describing the different roles of management levels from frontline management up to CEO.
• Fearless Change – not just for managers but for anyone trying to drive change, a set of helpful ‘design patterns’ to use when trying to make change happen.
• Never split the difference – on negotiation from a former FBI hostage negotiator.
• What got you here won’t get you there – dispelling the myth that quietly getting on and doing good work will get you to the next level.

Let me know in the comments if you have other book recommendations for anyone new to management.

Data science on the command line

There are just a few linux command line tools that I use many times a day!

less for checking the contents of files, verifying you’ve got the right output or input format, quickly examining data

grep for searching within files, especially as you can search for regular expressions

awk is incredibly useful for doing basic operations on text files, simple transformations from one format to another, or getting simple stats

Combine these with a few useful helpers like paste, diff, sort, wc, head, tail and cut, and you can do some really complex operations on actually quite large files.

Science and engineering gifts for young kids: 2014 edition

I wrote about science and engineering gifts last year, but now we’re all a year older and wiser it’s time to look around for more ideas.I’ve found that there are plenty of science and construction kits around for older kids, but durable pre-school friendly kit isn’t as common. My daughter has left the toddler years behind, and my son (now 18 months old) occasionally plays with the contents of the box, instead of just the box itself. So I’m covering both the toddler and pre-school ages, roughly 1.5-4 years.

Hence, here’s my 2014 xmas gift idea list; science and engineering gifts for pre-schoolers:

• Melissa and Doug construction set, for toddler engineers: blocks and bolts and even a screwdriver
• You are Stardust explaining to kids, with beautiful illustrations, how we all originate from stardust
• Astronaut duvet set for wannabe astronauts to dream of their journey to the stars
• Manhattan toy microscope a durable wooden microscope with different lenses and ‘petri dishes’ to look at
• Aquabot a robot fish, perfect for bathtime!
• Lunar explorer kit all the bits to recreate your favourite space mission
• 3d space cookie cutters, what pre-schooler doesn’t love to bake cookies?!
• Archiville, a city to build, and rebuild, any way you want
• Kids outdoor adventurer kit complete with insect hotel to build, bug book, and chalks and pencils for recording their adventures